Are there guiding principles as to which activation functions suit a given RL algorithm?

Question

Are there rules of thumb as to which activation functions work well (or which one would not) on the policy and value network of a class of RL algorithms? For hidden layers and for the output layer.

For example, I came across [1], which mentions ELU to be indispensable to MPO [2], and tanh (output activation) to be indispensable to SAC's Gaussian policy.